Saturday, December 31, 2011

SaaS (Software as a Service) and Multi-Tenancy

SaaS
SaaS can be defined as:
"software deployed as a hosted service and provided over the internet". 


   It is becoming an increasingly popular word, one that Software Vendors can ill-afford to ignore. It is opening up markets and segments that were previously inaccessible to vendors but it is also throwing up challenges like never before. 
    SaaS allows an application to scale to (theoretically) unlimited number of customers. It is typically achieved by on-demand horizontal and vertical scaling behind the scenes to provide a seamless experience to the user. 
    The practice of Application architecture has to adjust to account for SaaS and to Architect and Design applications for SaaS. Support for Multi-Tenancy is the approach some SaaS vendors follow to effectively address the SaaS challenges and this article will attempt to address the different aspects of Multi-Tenancy. So...


What is Multi-Tenancy
Multitenancy refers to a principle in software architecture where a single instance of the software runs on a server, serving multiple client organizations (tenants).


Why Multi-tenancy?
  • Great and Easy deployment. There is a single system to manage.
  • Cost savings associated with less hardware(or Virtualized Hardware)
  • Less software licenses to buy.

Multi-Tenancy Issues
  • Security testing has to be extensively done.
  • Expensive hardware to buy. Increased cost associated with big rack hardware.
  • Not easy to convert an existing application/software to Multi-tenant. 
  • Requires schema change to existing apps. 
  • Pure Multi-tenancy requires applications and infrastructure to scale-up to address demand.
  • Single point of failure issues. Which is the weakest link?
  • Multiple tenant metadata is difficult to manage.
  • Down time issues. What happens if the Database has to be brought down for maintenance?
  • Access control. This could be managed and configured at diff customers and less control over issues surrounding this.
  • Extensions to data model are tricky and sometimes impact all customers.
  • What about architecture that relies on communication between components in SaaS world and ones behind the firewall ?

Muti-tenancy versus Virtualization
Virtualization to a limited extent seems to be a good alternative for Multi-tenancy. However there are drawbacks. Would virtualization be profitable if there were 1000 customers deployments? How would you deploy and manage such an environment? Also what would we do if the Customers are "on-demand"? Which means they may want to use the system for brief periods of time and then be gone for days/weeks/months. How do you keep a Virtualization solution profitable?


But this is not to say that it just can't be done profitably. There are solutions out there that are using Virtualization to achieve Multi-Tenancy especially If the customer deployments are in single or very low double digits.Virtualization might (better) provide the following benefits:

  • Data Isolation.
  • Security. Both at the PaaS layer and at the Virtualized layer.
  • Performance ( one client can't directly impact the other's performance).

Designing for Multi-tenancy


Key points of Multi-Tenancy
  • Flexibility
  • Share-ability
  • Maintainability
  • Customizability

Architectural Constraints
  • Maintain a single code base to ease deployments and upgrades.
  • Share the data resources to have a consistent view of the Schema.
  • Components must be customizable at every possible level.
  • The Application Tier must be as stateless as possible to allow Scalability.

Trade offs
  • Complexity versus Time to market. What do the customers want and when?
  • Resource sharing vs Security/Availability. Who is my customer? Legal or SLA considerations?
  • Customize-ability vs Maintainability. A myriad of customizable options. Which one was chosen when this issue occurred? How do we fix this without affecting everyone else?

Interfaces
Multi-Tenant applications should expose (and consume) standard based interfaces like
  • REST
  • WS-*
Configurable
The application has to be configurable. In a traditional MVC style of application, the following would have to be extensively configurable:
Model: Allow schema extensions.
Controller: Allow new business logic to be plugged in or existing ones to be enhanced. Allow modification/customization of security policy.
View: Allow look and feel changes, changing of display items, screen order, messages etc. This does assume that there is a metadata somewhere that allows the configuration options and allows the particular view to be "instantiated" based on the metadata.



Security
Security includes, secure the SaaS model as whole and has an application level security architecture and a Data level security architecture. Data level Security is addressed in the next section. Application level security could involve either storing all user accounts with the SaaS provider and/or federating authentication to  trusted STSs ( Security Token Services) or trusted Identity Providers.
The SaaS application can also provide configurable Identity Management modules that either do the authentication/authorization themselves or federate as noted above. If the Authentication is federated, then we need some kind of a mapping service to map roles from Trusted servers to the roles/policies defined in the SaaS application.


Multi-Tenant Data Architecture
  • Separate Database 
    • Easy to Maintain
    • Customizable ( with probable issues later on)
    • Secure
    • Upgradable
    • Higher costs
  • Shared Database with separate Schema
    • Easier to Maintain
    • Customizable ( with probable issues later on)
    • More Secure
    • Upgradable but complex
    • lower costs
  • (Truly) Shared Database. This is the ideal scenario and the one most touted by purists. However the risk is that the different tenant data could inadvertently be mixed resulting in legal, compliance or contract issues. Also there are issues around Database sizes and multiple strategies including Partitioning and Segmentation have to be used to manage the Database. We won't go into details as there is more than adequate literature around these topics.
    • Complex Upgrade process
    • Impacts Multiple customers
    • Data Security at Data Access Layer.
    • Low Cost ( not assuming cost of design, development etc).


Which approach to choose?
Chose the Shared Database option if the number of tenants are higher to justify initial investment. However if number of users per tenant or database size per tenant or per tenant customizations are greater then choose the isolated model. 



Data-Security
Security strategies for shared database could include having views that filter the data visible to a tenant, using access control to the database itself and encryption at the data layer ( decryption at the DAC using tenant specific keys). 

Data-Extensibility
Multiple options are available including using name-value pairs to store data. Traditional data-extensibility approaches include have predefined fields and allowing extensions through metadata tables.

Data-Scalability
Data partitioning as noted in the above sections can aid in horizontal scaling of the Database.



Refer Multi-Tenant Data Architecture for further details.


Conclusion
Achieving the most optimum "degree" of multi-tenancy is something that the organization has to strive for. It could start with the Multi-Tenancy at the Infrastructure layer (IaaS), and the Platform layer (PaaS), move on to SaaS clusters that can provide some degree of Multi-Tenancy and then finally move on to the complete Multi-Tenancy at the Software layer (SaaS). 


Resources
  1. Wikipedia- Multitenancy
  2. Many degrees of Multi-tenancy is an excellent blog post that outlines the current debate and approaches for Multi-tenancy. 
  3. Multi-Tenant Data Architecture is an excellant resource that talks about Data design patterns for Multi-Tenancy.
The opinions and statements in this communication are my own and do not necessarily reflect the opinions or policies of CA.

No comments:

Post a Comment